A Method for Determining Optimal Leaf Picking Amount of Litchi Based on YOLOv7

Automatic picking of litchi is an important means to effectively improve the yield and quality of litchi. At present, branch and leaf occlusion is the critical factor that makes the picking robot unable to locate and pick fruit accurately. In this paper, a method based on YOLOv7 to calculate the optimal leaf-picking amount is proposed to optimize the problem. First of all, three groups of litchi trees were treated with different leaf-picking treatments at flowering stage. Then, the images of litchi s’ fruits and leaves are collected regularly to build a data set to train the YOLOv7 model. Finally, it comes to a conclusion that the optimal leaf-picking amount at litchi flowering stage was determined, by comparing the fruit recognition accuracy and fruit data under different leaf-picking amounts. According to the litchi test conclusions, the model’s AP values under the three situations of no picking, three pairs of leaves, and six pairs of leaves were 30.16%, 70.62%, and 82.75%, respectively. In addition, the yield of single litchi tree were 30kg, 37kg, 35kg. Therefore, considering the accuracy of the algorithm and the fruit data, the optimal leaf picking quantity is determined as 3 pairs of leaves. The results show that the method is feasible and can provide information support for litchi production management.